Runoff and sediment yield modeling in data-sparse catchments in the Garehsoo River basin, northern Iran

To improve management and conservation in ungauged catchments in northern Iran, we developed regional models to predict annual runoff and discharge using multi-year time series from 20 gauged catchments in Ardabil, Province. We employed correlation, Ward cluster analysis, and principal component analysis to reduce the total parameter set (69 parameters represent catchment geometry, geology, soil, rainfall, and climate) to robust sets for regression analysis. The reduction method based on PCA using non-log-transformed data produced the best results, based on several error metrics (modified Nash Sutcliffe efficiency, R², root mean squared error, and bias). Important parameters included area and mean annual precipitation, as well as those indicative of catchment shape and geology. Despite excellent ability to simulate mean annual discharge and sediment yield in the 20 catchments, the models performed poorly when applied to estimating year-to-year discharge and sediment yield in individual catchments. Poor prediction likely resulted from (a) rainfall data of insufficient spatial resolution in the large area with greatly varying elevation including mountains; and (b) inability to include data reflecting yearly changes in land-cover/land-use. Nevertheless, the regional models are likely useful for estimating mean runoff and sediment yields in other catchments in the region with similar geology, geomorphology, and climate.